2 * Copyright (C) 2014 Linaro Ltd. <ard.biesheuvel@linaro.org>
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
10 #include <linux/kernel.h>
11 #include <linux/module.h>
12 #include <linux/sort.h>
14 #include <asm/cache.h>
15 #include <asm/opcodes.h>
17 #define PLT_ENT_STRIDE L1_CACHE_BYTES
18 #define PLT_ENT_COUNT (PLT_ENT_STRIDE / sizeof(u32))
19 #define PLT_ENT_SIZE (sizeof(struct plt_entries) / PLT_ENT_COUNT)
21 #ifdef CONFIG_THUMB2_KERNEL
22 #define PLT_ENT_LDR __opcode_to_mem_thumb32(0xf8dff000 | \
25 #define PLT_ENT_LDR __opcode_to_mem_arm(0xe59ff000 | \
30 u32 ldr
[PLT_ENT_COUNT
];
31 u32 lit
[PLT_ENT_COUNT
];
34 u32
get_module_plt(struct module
*mod
, unsigned long loc
, Elf32_Addr val
)
36 struct plt_entries
*plt
= (struct plt_entries
*)mod
->arch
.plt
->sh_addr
;
40 * Look for an existing entry pointing to 'val'. Given that the
41 * relocations are sorted, this will be the last entry we allocated.
44 if (mod
->arch
.plt_count
> 0) {
45 plt
+= (mod
->arch
.plt_count
- 1) / PLT_ENT_COUNT
;
46 idx
= (mod
->arch
.plt_count
- 1) % PLT_ENT_COUNT
;
48 if (plt
->lit
[idx
] == val
)
49 return (u32
)&plt
->ldr
[idx
];
51 idx
= (idx
+ 1) % PLT_ENT_COUNT
;
56 mod
->arch
.plt_count
++;
57 BUG_ON(mod
->arch
.plt_count
* PLT_ENT_SIZE
> mod
->arch
.plt
->sh_size
);
60 /* Populate a new set of entries */
61 *plt
= (struct plt_entries
){
62 { [0 ... PLT_ENT_COUNT
- 1] = PLT_ENT_LDR
, },
68 return (u32
)&plt
->ldr
[idx
];
71 #define cmp_3way(a,b) ((a) < (b) ? -1 : (a) > (b))
73 static int cmp_rel(const void *a
, const void *b
)
75 const Elf32_Rel
*x
= a
, *y
= b
;
78 /* sort by type and symbol index */
79 i
= cmp_3way(ELF32_R_TYPE(x
->r_info
), ELF32_R_TYPE(y
->r_info
));
81 i
= cmp_3way(ELF32_R_SYM(x
->r_info
), ELF32_R_SYM(y
->r_info
));
85 static bool is_zero_addend_relocation(Elf32_Addr base
, const Elf32_Rel
*rel
)
87 u32
*tval
= (u32
*)(base
+ rel
->r_offset
);
90 * Do a bitwise compare on the raw addend rather than fully decoding
91 * the offset and doing an arithmetic comparison.
92 * Note that a zero-addend jump/call relocation is encoded taking the
93 * PC bias into account, i.e., -8 for ARM and -4 for Thumb2.
95 switch (ELF32_R_TYPE(rel
->r_info
)) {
99 case R_ARM_THM_JUMP24
:
100 upper
= __mem_to_opcode_thumb16(((u16
*)tval
)[0]);
101 lower
= __mem_to_opcode_thumb16(((u16
*)tval
)[1]);
103 return (upper
& 0x7ff) == 0x7ff && (lower
& 0x2fff) == 0x2ffe;
108 return (__mem_to_opcode_arm(*tval
) & 0xffffff) == 0xfffffe;
113 static bool duplicate_rel(Elf32_Addr base
, const Elf32_Rel
*rel
, int num
)
115 const Elf32_Rel
*prev
;
118 * Entries are sorted by type and symbol index. That means that,
119 * if a duplicate entry exists, it must be in the preceding
125 prev
= rel
+ num
- 1;
126 return cmp_rel(rel
+ num
, prev
) == 0 &&
127 is_zero_addend_relocation(base
, prev
);
130 /* Count how many PLT entries we may need */
131 static unsigned int count_plts(const Elf32_Sym
*syms
, Elf32_Addr base
,
132 const Elf32_Rel
*rel
, int num
)
134 unsigned int ret
= 0;
138 for (i
= 0; i
< num
; i
++) {
139 switch (ELF32_R_TYPE(rel
[i
].r_info
)) {
144 case R_ARM_THM_JUMP24
:
146 * We only have to consider branch targets that resolve
147 * to undefined symbols. This is not simply a heuristic,
148 * it is a fundamental limitation, since the PLT itself
149 * is part of the module, and needs to be within range
150 * as well, so modules can never grow beyond that limit.
152 s
= syms
+ ELF32_R_SYM(rel
[i
].r_info
);
153 if (s
->st_shndx
!= SHN_UNDEF
)
157 * Jump relocations with non-zero addends against
158 * undefined symbols are supported by the ELF spec, but
159 * do not occur in practice (e.g., 'jump n bytes past
160 * the entry point of undefined function symbol f').
161 * So we need to support them, but there is no need to
162 * take them into consideration when trying to optimize
163 * this code. So let's only check for duplicates when
164 * the addend is zero.
166 if (!is_zero_addend_relocation(base
, rel
+ i
) ||
167 !duplicate_rel(base
, rel
, i
))
174 int module_frob_arch_sections(Elf_Ehdr
*ehdr
, Elf_Shdr
*sechdrs
,
175 char *secstrings
, struct module
*mod
)
177 unsigned long plts
= 0;
178 Elf32_Shdr
*s
, *sechdrs_end
= sechdrs
+ ehdr
->e_shnum
;
179 Elf32_Sym
*syms
= NULL
;
182 * To store the PLTs, we expand the .text section for core module code
183 * and for initialization code.
185 for (s
= sechdrs
; s
< sechdrs_end
; ++s
) {
186 if (strcmp(".plt", secstrings
+ s
->sh_name
) == 0)
188 else if (s
->sh_type
== SHT_SYMTAB
)
189 syms
= (Elf32_Sym
*)s
->sh_addr
;
192 if (!mod
->arch
.plt
) {
193 pr_err("%s: module PLT section missing\n", mod
->name
);
197 pr_err("%s: module symtab section missing\n", mod
->name
);
201 for (s
= sechdrs
+ 1; s
< sechdrs_end
; ++s
) {
202 Elf32_Rel
*rels
= (void *)ehdr
+ s
->sh_offset
;
203 int numrels
= s
->sh_size
/ sizeof(Elf32_Rel
);
204 Elf32_Shdr
*dstsec
= sechdrs
+ s
->sh_info
;
206 if (s
->sh_type
!= SHT_REL
)
209 /* ignore relocations that operate on non-exec sections */
210 if (!(dstsec
->sh_flags
& SHF_EXECINSTR
))
213 /* sort by type and symbol index */
214 sort(rels
, numrels
, sizeof(Elf32_Rel
), cmp_rel
, NULL
);
216 plts
+= count_plts(syms
, dstsec
->sh_addr
, rels
, numrels
);
219 mod
->arch
.plt
->sh_type
= SHT_NOBITS
;
220 mod
->arch
.plt
->sh_flags
= SHF_EXECINSTR
| SHF_ALLOC
;
221 mod
->arch
.plt
->sh_addralign
= L1_CACHE_BYTES
;
222 mod
->arch
.plt
->sh_size
= round_up(plts
* PLT_ENT_SIZE
,
223 sizeof(struct plt_entries
));
224 mod
->arch
.plt_count
= 0;
226 pr_debug("%s: plt=%x\n", __func__
, mod
->arch
.plt
->sh_size
);
This page took 0.036392 seconds and 5 git commands to generate.